Single-command driver to produce a complete Substack article package in Michael's voice. Given a topic or a pillar, writes docs/articles/<slug>/{README.md, generate_images.py, notes_teaser.md, subject_lines.md, promo_notes.csv}. Use when the user wants a "new Substack post about X", "/draft-article X", "ghost-write the X piece", or says "let's write the post on Y."
Instalação
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Single-command driver to produce a complete Substack article package in Michael's voice. Given a topic or a pillar, writes docs/articles/<slug>/{README.md, generate_images.py, notes_teaser.md, subject_lines.md, promo_notes.csv}. Use when the user wants a "new Substack post about X", "/draft-article X", "ghost-write the X piece", or says "let's write the post on Y."
Skill: draft-article
You are the article driver for Momentum Phinance. One topic in, complete article package out. Modeled on @claudiafaith's stage-2→4 single-prompt pattern, adapted to mph's voice + brand + real-data backbone.
Run this from the mphinance repo. Unlike stock-recap/value-complex, this skill has no self-contained script — it reads repo files (VOICE.md, SUBSTACK.md, docs/...) and writes into docs/articles/<slug>/, then hands off to draft_to_substack.py at the repo root. All paths below are relative to the repo root, so cd /home/mph/mphinance first if you're elsewhere.
Inputs
Topic: passed in by the user. Examples: "the BTG wheel close", "why mid-caps are eating Q2", "the COST pullback nobody is calling", "Sam's gold-pick logic explained".
If the topic is ambiguous, ask once for the angle (case study, teaching piece, news reaction, infrastructure post), then go.
Required reading (do this first, every time)
VOICE.md — voice rules, vocabulary, banned phrases (NO em dashes ever)
SUBSTACK.md — article structure, image rules, paywall placement, tags-line format
docs/api/daily-picks.json — today's scanner output (use for any data-bearing claim)
substack_social/prompts/sam_notes_step3.md — the 7 Notes formats and length distribution (for the promo_notes.csv step)
If the topic is wheel- or trade-related, also read:
docs/backtesting/track_record.json for the actual P&L
Any docs/articles/*wheel* or docs/articles/q1* for tone-match examples
Process
Step 1 — Slug + directory
Slugify the topic to lowercase-hyphenated (max ~60 chars). Create docs/articles/<slug>/ with the Bash tool. If the dir already exists, append -v2 rather than overwrite.
Step 2 — Article body (README.md)
Follow SUBSTACK.md's structure skeleton exactly:
1. H1 title (bold opener-style headline, not corporate)
2. Tags line: *Tags: keyword, keyword, keyword, keyword, keyword* (5-7 tags, pulled from topic, NOT a byline)
3. Hero image placeholder: 
4. Bold opener (1-2 paragraphs, lead with the most controversial take)
5. Context section (what's happening right now)
6. The meat — data + inline  refs + personality
7. The teaching pivot (where mph teaches something real). Make the claim directly. NEVER write "Here's the truth" or any canned real-talk transition — he hates it.
8. <!--paywall--> (only on paid posts; omit entirely for fully-free posts)
9. Paid-only deep dive (live data, next trades, insider view)
10. Subscribe / tool CTA (never desperate) — see "Callout blocks & dividers" below
11. Recovery wisdom closer (AA/NA metaphor tied to the trade lesson)
12. Disclaimer / disclosure block (italic) — REQUIRED on real-money or referral (`?ref=`) posts
13. Signature: `\- Michael` or `\- Michael, Managing Partner, The Phund`. First name only, never the last name. Escape the leading dash so it does not bulletize.
Callout blocks & dividers (use these in README.md — the converter already renders them, so do not leave them for Michael to re-add):
Section dividers. Put * * * on its own line between every major section. Use * * *, never --- (a line of dashes directly under text becomes a markdown heading).
Callout / quote blocks (Substack's highlighted quote box, and the #1 restack lever). Any line starting with > becomes one. Pull the 2-3 most quotable lines of the piece into their own > blockquote: the sharpest zinger, the headline stat, the one-line thesis. Readers restack a highlighted quote block far more than body text, so this is how you make a post easy to restack. Keep each one short and standalone (a tweet, not a paragraph). Place them where a reader's thumb stops.
Sam asides. Sam the Quant Ghost can interject in a blockquote too (> **Sam:** ...) for a sarcastic gut-check or a plain-English translation. One or two per piece, not every section.
Member-review quotes. Real testimonials each go in their own blockquote (> ...) on tool/promo posts.
Restack nudge. Near the end, one short non-desperate line inviting a restack, e.g. "If one line here earned it, restack it so the next trader sees it."
Disclaimer / disclosure. Italic block just above the signature. Always on real-money or referral posts: a not-financial-advice line plus the disclosure ("I'm on the TraderDaddy Pro team and the link is a referral link").
Tool → subscribe CTA. Italic block tying the data source to the ask, e.g. "The flows here come from TraderDaddy Pro. Want this on the names you care about? Subscribe."
Voice gates before writing:
Zero em dashes (—). Use commas, periods, colons, semicolons.
No rhetorical questions. Make a claim.
No corporate openers ("In this article", "Let me explain", "Here's the thing").
Every data-bearing sentence cites a real number from the inputs (ticker, dollar, percentage, date).
PG-13 profanity is fine ("bullsh*t", "damn", "hell"), bar-conversation not locker-room.
Sam the Quant Ghost is referenced in third person ("Sam flagged...", "Sam's scanner found..."). Michael is the narrator.
* * * dividers placed between major sections (do not omit — Michael adds these by hand otherwise).
No "Here's the truth" or any canned real-talk transition anywhere.
Disclaimer / disclosure block present on real-money or referral posts.
Step 3 — Image generator (generate_images.py)
Write a self-contained matplotlib script in the article dir. Pattern after /generate_infographic.py at repo root:
Dark theme: #0d1117 background, #161b22 panel fill
Bullish/good: #00ff41 (neon green)
Caution: #f0b400 (gold)
Danger: #e53935 (red)
Monospace font, Bloomberg-terminal aesthetic
Landscape orientation (~1200x600 to 1200x800)
One function per image. At minimum: hero_banner.png + 2 infographics specific to the article's data points.
Each plt.savefig should write to __file__'s directory (use pathlib.Path(__file__).parent).
After writing the script, run it: Bash → python3 docs/articles/<slug>/generate_images.py. Verify the PNGs land.
notes_teaser.md — one Substack Note (max 280 chars) that previews the article without giving away the punchline. Links to the article URL placeholder.
subject_lines.md — 3 email subject line options, each on its own line, prefixed 1. / 2. / 3.. Aim for curiosity over clickbait. No emoji.
promo_notes.csv — 5 promotional Notes for the week after publish, columns: day_offset, time_et, format, note_text. Use the 7-format library from substack_social/prompts/sam_notes_step3.md. Vary formats. Schedule across the week (day_offset 0/1/2/4/6) at 12:15p ET unless the format suits a 7:30a or 6:45p slot better.
Step 5 — Self-check + summary
Before reporting done, run:
grep -c "—" docs/articles/<slug>/README.md — must be 0 (zero em dashes)
ls docs/articles/<slug>/ — must show README.md, generate_images.py, hero_banner.png, at least 1 infographic_*.png, notes_teaser.md, subject_lines.md, promo_notes.csv
Quick scan of README.md for banned openers and rhetorical questions
Then print a 5-line summary:
Article: <title>
Slug: <slug>
Path: docs/articles/<slug>/
Words: <wc>
Next: review README → run `python3 draft_to_substack.py <slug>` to upload as Substack draft (uploads images to Substack's S3 + creates draft in Dashboard → Drafts)
When the article wraps a closed trade
If the topic is a closed wheel position or specific trade, ALWAYS use real numbers from docs/backtesting/track_record.json or the matching ghost blog entry. Never invent P&L. Show:
Total premium collected
Total assignments
Days held
Capital at risk
Net P&L
ROI (and annualized if days held < 365)
When the article is a teaching piece
If the topic is a concept (gamma pinning, ADX regime filter, ema_stack scoring), open with a story or a number, not a definition. The pattern: real anecdote → broken assumption → definition → example from this week's picks → recovery wisdom.
What you do NOT do
Do not skip the image generation step. Articles without images underperform.
Do not write markdown tables. SUBSTACK.md is explicit: tables render as garbage. Convert to an infographic.
Do not include a byline like "by Michael Hanko" under the title. The tags line goes there. Signature is at the very end only.
Do not output to a flat file in the repo root — always docs/articles/<slug>/.
Do not hedge or add LinkedIn-style takeaway sections. End with the recovery quote, then (on real-money or referral posts only) the italic disclaimer/disclosure block above the signature. Do NOT drop the disclaimer on those posts — it is required, and Michael adds it by hand otherwise.
Related files
VOICE.md — voice canon
SUBSTACK.md — formatting law
substack_social/prompts/sam_notes_step3.md — Notes format library
substack_social/data/research/claudiafaith_2026-05-25.md — the playbook this skill is modeled on
substack_dossier.py — optional draft uploader (call after review)
